Discover millions of ebooks, audiobooks, and so much more with a free trial

Only $11.99/month after trial. Cancel anytime.

Handbook of Financial Markets: Dynamics and Evolution
Handbook of Financial Markets: Dynamics and Evolution
Handbook of Financial Markets: Dynamics and Evolution
Ebook1,308 pages31 hours

Handbook of Financial Markets: Dynamics and Evolution

Rating: 0 out of 5 stars

()

Read preview

About this ebook

The models of portfolio selection and asset price dynamics in this volume seek to explain the market dynamics of asset prices. Presenting a range of analytical, empirical, and numerical techniques as well as several different modeling approaches, the authors depict the state of debate on the market selection hypothesis. By explicitly assuming the heterogeneity of investors, they present models that are descriptive and normative as well, making the volume useful for both finance theorists and financial practitioners.

* Explains the market dynamics of asset prices, offering insights about asset management approaches
* Assumes a heterogeneity of investors that yields descriptive and normative models of portfolio selections and asset pricing dynamics
LanguageEnglish
Release dateJun 12, 2009
ISBN9780080921433
Handbook of Financial Markets: Dynamics and Evolution

Related to Handbook of Financial Markets

Titles in the series (8)

View More

Related ebooks

Personal Finance For You

View More

Related articles

Reviews for Handbook of Financial Markets

Rating: 0 out of 5 stars
0 ratings

0 ratings0 reviews

What did you think?

Tap to rate

Review must be at least 10 words

    Book preview

    Handbook of Financial Markets - Elsevier Science

    Schenk-Hoppé

    Brief Table of Contents

    Copyright

    List of Contributors

    Preface

    Introduction to the Series

    Chapter 1. Thought and Behavior Contagion in Capital Markets

    Chapter 2. How Markets Slowly Digest Changes in Supply and Demand

    Chapter 3. Stochastic Behavioral Asset-Pricing Models and the Stylized Facts

    Chapter 4. Complex Evolutionary Systems in Behavioral Finance

    Chapter 5. Heterogeneity, Market Mechanisms, and Asset Price Dynamics

    Chapter 6. Perfect Forecasting, Behavioral Heterogeneities, and Asset Prices

    Chapter 7. Market Selection and Asset Pricing

    Chapter 8. Rational Diverse Beliefs and Market Volatility

    Chapter 9. Evolutionary Finance

    Table of Contents

    Copyright

    List of Contributors

    Preface

    Introduction to the Series

    Chapter 1. Thought and Behavior Contagion in Capital Markets

    1.1.. INTRODUCTION

    1.2.. SOURCES OF BEHAVIORAL CONVERGENCE

    1.3.. RATIONAL LEARNING AND INFORMATION CASCADES: BASIC IMPLICATIONS

    1.4.. WHAT IS COMMUNICATED OR OBSERVED?

    1.4.1.. Observation of Past Actions Only

    1.4.2.. Observation of Consequences of Past Actions

    1.4.3.. Conversation, Media, and Advertising

    1.5.. PSYCHOLOGICAL BIAS

    1.6.. REPUTATION, CONTRACTS, AND HERDING

    1.7.. SECURITY ANALYSIS

    1.7.1.. Investigative Herding

    1.7.2.. Herd Behavior by Stock Analysts and Other Forecasters

    1.8.. HERD BEHAVIOR AND CASCADES IN SECURITY TRADING

    1.8.1.. Evidence on Herding in Securities Trades

    1.8.2.. Financial Market Runs and Contagion

    1.8.3.. Exploiting Herding and Cascades

    1.9.. MARKETS, EQUILIBRIUM PRICES, AND BUBBLES

    1.10.. CASCADES AND HERDING IN FIRM BEHAVIOR

    1.10.1.. Investment and Financing Decisions

    1.10.2.. Disclosure and Reporting Decisions

    1.11.. CONTAGION OF FINANCIAL MEMES

    1.12.. CONCLUSION

    Chapter 2. How Markets Slowly Digest Changes in Supply and Demand

    2.1.. INTRODUCTION

    2.1.1.. Overview

    2.1.2.. Organization

    2.1.3.. Motivation and Scope

    2.1.4.. Approach to Model Building

    2.2.. MARKET STRUCTURE

    2.3.. INFORMATION, LIQUIDITY, AND EFFICIENCY

    2.3.1.. Information and Fundamental Values

    2.3.2.. Market Efficiency

    2.3.3.. Trading and Information

    2.3.4.. Different Explanations for Market Impact

    2.3.5.. Noise Trader Models and Informed vs. Uninformed Trading

    2.3.6.. A Critique of the Noise Trader Explanation of Market Impact

    2.3.7.. The Liquidity Paradox: Prices Are Not in Equilibrium

    2.3.8.. Time Scales and Market Ecology

    2.3.9.. The Volatility Puzzle

    2.3.10.. The Kyle Model

    2.4.. LARGE FLUCTUATIONS AND LONG MEMORY OF ORDER FLOW

    2.4.1.. Empirical Evidence for Long Memory of Order Flow

    2.4.2.. On the Origin of Long Memory of Order Flow

    2.4.3.. Theory for Long Memory in Order Flow Based on Strategic Order Splitting

    2.4.4.. Evidence Based on Exchange Membership Codes

    2.4.5.. Evidence for Heavy Tails in Volume

    2.5.. SUMMARY OF EMPIRICAL RESULTS FOR DIVERSE TYPES OF MARKET IMPACT

    2.5.1.. Impact of Individual Transactions

    2.5.2.. Impact of Aggregate Transactions

    2.5.3.. Hidden Order Impact

    2.5.4.. Upstairs Market Impact

    2.6.. THEORY OF MARKET IMPACT

    2.6.1.. Why Is Individual Transaction Impact Concave?

    2.6.2.. A Fixed Permanent Impact Model

    2.6.3.. The MRR Model

    2.6.4.. A Transient Impact Framework

    2.6.5.. History Dependent, Permanent Impact

    2.6.6.. Empirical Results

    2.6.7.. Impact of a Large Hidden Order

    2.6.8.. Aggregated Impact

    2.7.. THE DETERMINANTS OF THE BID–ASK SPREAD

    2.7.1.. The Basic Economics of Spread and Impact

    2.7.2.. Models for the Bid–Ask Spread

    2.7.3.. Limit vs. Market Orders: The Microstructure Phase Diagram

    2.7.4.. Spread Dynamics After a Temporary Liquidity Crisis

    2.8.. LIQUIDITY AND VOLATILITY

    2.8.1.. Liquidity and Large Price Changes

    2.8.2.. Volume vs. Liquidity Fluctuations as Proximate Causes of Volatility

    2.8.3.. Spread vs. Volatility

    2.8.4.. Market Cap Effects

    2.9.. ORDER BOOK DYNAMICS

    2.9.1.. Heavy Tails in Order Placement and the Shape of the Order Book

    2.9.2.. Volume at Best Prices: The Glosten-Sandas Model

    2.9.3.. Statistical Models of Order Flow and Order Books

    2.10.. IMPACT AND OPTIMIZED EXECUTION STRATEGIES

    2.11.. TOWARD AN EMPIRICAL CHARACTERIZATION OF A MARKET ECOLOGY

    2.11.1.. Identifying Hidden Orders

    2.11.2.. Specialization of Strategies

    2.12.. CONCLUSION

    APPENDIX 2.1. MECHANICAL VS. NONMECHANICAL IMPACT

    A2.1.1.. Definition of Mechanical Impact for Order Books

    A2.1.2.. Empirical Results

    APPENDIX 2.2. VOLUME FLUCTUATIONS

    APPENDIX 2.3. THE BID–ASK SPREAD IN THE MRR MODEL

    Chapter 3. Stochastic Behavioral Asset-Pricing Models and the Stylized Facts

    3.1.. INTRODUCTION

    3.2.. THE STYLIZED FACTS OF FINANCIAL DATA

    3.2.1.. Martingales, Lack of Predictability, and Informational Efficiency

    3.2.2.. Fat Tails of Asset Returns

    3.2.3.. Volatility Clustering and Dependency in Higher Moments

    3.2.4.. Other Stylized Facts

    3.3.. THE STYLIZED FACTS AS SCALING LAWS

    3.4.. BEHAVIORAL ASSET-PRICING MODELS WITH INTERACTING AGENTS

    3.4.1.. Interaction of Chartists and Fundamentalists and Nonlinear Dynamics of Asset Prices

    3.4.2.. Kirman's Model of Opinion Formation and Speculation

    3.4.3.. Beyond Local Interactions: Socioeconomic Group Dynamics in Financial Markets

    3.4.4.. Lattice Topologies of Agents' Connections

    3.5.. CONCLUSION

    Chapter 4. Complex Evolutionary Systems in Behavioral Finance

    4.1.. INTRODUCTION

    4.2.. AN ASSET-PRICING MODEL WITH HETEROGENEOUS BELIEFS

    4.2.1.. The Fundamental Benchmark with Rational Agents

    4.2.2.. Heterogeneous Beliefs

    4.2.3.. Evolutionary Dynamics

    4.2.4.. Forecasting Rules

    4.3.. SIMPLE EXAMPLES

    4.3.1.. Costly Fundamentalists vs. Trend Followers

    4.3.2.. Fundamentalists vs. Opposite Biases

    4.3.3.. Fundamentalists vs. Trend and Bias

    4.3.4.. Efficiency

    4.3.5.. Wealth Accumulation

    4.3.6.. Extensions

    4.4.. MANY TRADER TYPES

    4.5.. EMPIRICAL VALIDATION

    4.5.1.. The Model in Price-to-Cash Flows

    4.5.2.. Estimation of a Simple Two-Type Example

    4.5.3.. Empirical Implications

    4.6.. LABORATORY EXPERIMENTS

    4.6.1.. Learning to Forecast Experiments

    4.6.2.. The Price-Generating Mechanism

    4.6.3.. Benchmark Expectations Rules

    4.6.4.. Aggregate Behavior

    4.6.5.. Individual Prediction Strategies

    4.6.6.. Profitability

    4.7.. CONCLUSION

    APPENDIX 4.1. BIFURCATION THEORY

    A4.1.1.. Basic Concepts from Dynamical Systems

    APPENDIX 4.2. BIFURCATION SCENARIOS

    A4.2.1.. The Saddle-Node Bifurcation

    A4.2.2.. The Period-Doubling Bifurcation

    A4.2.3.. The Hopf Bifurcation

    A4.2.4.. The Pitchfork Bifurcation

    Chapter 5. Heterogeneity, Market Mechanisms, and Asset Price Dynamics

    5.1.. INTRODUCTION

    5.2.. HETEROGENEITY AND MARKET-CLEARING MECHANISMS

    5.2.1.. Portfolio Optimization

    5.2.2.. Utility Functions

    5.2.3.. Market-Clearing Mechanisms

    5.2.4.. Noise

    5.2.5.. Expectations Feedback

    5.3.. PRICE DYNAMICS IMPLIED BY THE CARA UTILITY FUNCTION

    5.3.1.. Fundamental Price and the Optimal Demand

    5.3.2.. Formation of Heterogeneous Beliefs

    5.3.3.. Performance Measure and Switching

    5.3.4.. Price Behavior under the Walrasian Auctioneer Mechanism

    5.3.5.. Price Behavior under the Market-Maker Mechanism

    5.4.. PRICE BEHAVIOR AND WEALTH DYNAMICS IMPLIED BY THE CRRA UTILITY

    5.4.1.. Optimal Portfolio and Wealth Dynamics

    5.4.2.. Price and Wealth Behavior with a Walrasian Auctioneer

    5.4.3.. Price and Wealth Behavior with a Market Maker

    5.5.. EMPIRICAL BEHAVIOR

    5.5.1.. Stylized Facts in the S&P 500

    5.5.2.. A Market Fraction Model and Its Stylized Behavior

    5.5.3.. Econometric Characterization of the Power-Law Behavior

    5.6.. HETEROGENEITY IN A DYNAMIC MULTIASSET FRAMEWORK

    5.6.1.. Optimization of a Many Risky Asset Portfolio with Heterogeneous Beliefs

    5.6.2.. An Example of Two Risky Assets and Two Beliefs

    5.7.. THE CONTINUOUS STOCHASTIC DYNAMICS OF SPECULATIVE BEHAVIOR

    5.7.1.. Stochastic Models with Heterogeneous Beliefs

    5.7.2.. A Continuous Stochastic Model with Fundamentalists and Chartists

    5.7.3.. A Random Dynamical System and Stochastic Bifurcations

    5.8.. CONCLUSION

    Chapter 6. Perfect Forecasting, Behavioral Heterogeneities, and Asset Prices

    6.1.. INTRODUCTION

    6.2.. THE CAPM AS A TWO-PERIOD EQUILIBRIUM MODEL

    6.2.1.. Portfolio Selection with One Risk-Free Asset

    6.2.2.. Two-Fund Separation

    6.2.3.. Existence and Uniqueness of Equilibrium

    6.2.4.. Equilibria with Heterogeneous Beliefs

    6.3.. HETEROGENEOUS BELIEFS AND SOCIAL INTERACTION

    6.3.1.. Temporary Equilibria

    6.3.2.. Perfect Forecasting Rules

    6.3.3.. Systematic and Nonsystematic Risk

    6.3.4.. Selecting Mediators

    EXAMPLE 6.1

    EXAMPLE 6.2

    EXAMPLE 6.3

    6.3.5.. Dynamic Stability with Rational Expectations

    6.4.. MULTIPERIOD PLANNING HORIZONS

    6.4.1.. Overlapping Cohorts of Investors

    6.4.2.. Temporary Equilibria

    6.4.3.. Perfect Forecasting Rules

    Chapter 7. Market Selection and Asset Pricing

    7.1.. INTRODUCTION

    7.1.1.. Evolution in Biology and Economics

    7.1.2.. A Short History of the Market Selection Hypothesis

    7.1.3.. Scope of This Chapter

    7.2.. THE ECONOMY

    7.2.1.. Traders

    7.2.2.. Beliefs

    7.3.. EQUILIBRIUM ALLOCATIONS AND PRICES

    7.3.1.. Pareto Optimality

    7.3.2.. Competitive Equilibrium

    7.4.. SELECTION

    7.4.1.. Literature

    7.4.2.. A Leading Example

    7.4.3.. Selection in Complete IID Markets

    7.4.4.. The Basic Equations

    7.4.5.. Who Survives? Necessity

    7.4.6.. Selection and Market Equilibrium

    7.4.7.. More General Stochastic Processes

    7.5.. MULTIPLE SURVIVORS

    7.5.1.. Who Survives? Sufficiency

    7.6.. THE LIFE AND DEATH OF NOISE TRADERS

    7.6.1.. The Importance of Market Structure

    7.6.2.. Laws of Large Numbers

    7.7.. ROBUSTNESS

    7.7.1.. Unbounded Economies

    7.7.2.. Incomplete Markets

    7.7.3.. Differential Information

    7.7.4.. Selection over Non-EU Traders

    7.7.5.. Selection over Rules

    7.8.. CONCLUSION

    Chapter 8. Rational Diverse Beliefs and Market Volatility

    8.1.. INTRODUCTION

    8.2.. CAN MARKET DYNAMICS BE EXPLAINED BY ASYMMETRIC PRIVATE INFORMATION?

    8.2.1.. A General Model of Asset Pricing under Asymmetric Information

    8.2.2.. Dynamic Infinite Horizon Models

    8.2.3.. Is Asymmetric Information a Satisfactory Theory of Market Dynamics?

    8.3.. DIVERSE BELIEFS WITH COMMON INFORMATION: THE GENERAL THEORY

    8.3.1.. A Basic Principle: Rational Diversity Implies Volatility

    8.3.2.. Stability and Rationality in a General Nonstationary Economy

    EXAMPLE 8.1

    EXAMPLE 8.2

    8.3.3.. Belief Rationality and the Conditional Stability Theorem

    8.3.4.. Describing Individual and Market Beliefs with Markov State Variables

    8.3.5.. Asset Pricing with Heterogeneous Beliefs: An Illustrative Model and Implications

    8.4.. EXPLAINING MARKET DYNAMICS WITH SIMULATION MODELS OF DIVERSE BELIEFS

    8.4.1.. Introduction: On Simulation Methods and the Main Results

    8.4.2.. Anatomy of Market Volatility

    8.4.3.. Volatility of Foreign Exchange Rates and the Forward Discount Bias

    8.4.4.. Macroeconomic Applications

    8.5.. CONCLUSION AND OPEN PROBLEMS

    Chapter 9. Evolutionary Finance

    9.1.. INTRODUCTION

    9.1.1.. Motivation and Background

    9.1.2.. Applications and Real-World Implications

    9.1.3.. Structure of Chapter

    9.1.4.. Dynamics and Evolution

    9.1.5.. Horse Races and the Kelly Rule

    9.2.. EVOLUTIONARY MODELS OF FINANCIAL MARKETS

    9.2.1.. Components of the Models

    9.2.2.. Discussion of the Assumptions

    9.2.3.. Outline of the Dynamics

    9.3.. AN EVOLUTIONARY MODEL WITH SHORT-LIVED ASSETS

    9.3.1.. The Model

    9.3.2.. Analysis of Local Dynamics

    9.3.3.. An Example

    9.3.4.. The Generalized Kelly Rule

    9.3.5.. Global Dynamics with Adaptive Strategies

    9.4.. AN EVOLUTIONARY STOCK MARKET MODEL

    9.4.1.. Local Dynamics

    9.4.2.. Global Dynamics with Constant Strategies

    9.4.3.. Kelly Rule in General Equilibrium

    9.5.. APPLICATIONS

    9.5.1.. Simulation Studies

    9.5.2.. Dynamics of Strategies: Genetic Programming

    9.5.3.. Empirical Tests of Evolutionary Asset Pricing

    9.6.. CONTINUOUS-TIME EVOLUTIONARY FINANCE

    9.7.. CONCLUSION

    Copyright

    North-Holland is an imprint of Elsevier

    Radarweg 29, PO Box 211, 1000 AE

    Amsterdam, The Netherlands

    Linacre House, Jordan Hill

    Oxford, OX2 8DP, UK

    30 Corporate Drive, Suite 400

    Burlington, MA 01803, USA

    525 B Street, Suite 1900

    San Diego, California 92101-4495, USA

    Copyright © 2009, Elsevier Inc. All rights reserved.

    No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including recording, photocopying, or otherwise, without the prior written permission of the publisher.

    Permissions may be sought directly from Elsevie]r's Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; email: permissions@elsevier.com. Alternatively you can submit your request online by visiting the Elsevier Web site at www.elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material.

    Recognizing the importance of preserving what has been written, Elsevier prints its books on acid-free paper whenever possible.

    British Library Cataloguing in Publication Data

    A catalogue record for this book is available from the British Library.

    Library of Congress Cataloging-in-Publication Data

    Application submitted.

    ISBN: 978-0-12-374258-2

    For information on all Elsevier publications, visit our Web site at www.books.elsevier.com.

    Printed in the United States of America

    09 10 11 12 13 10 9 8 7 6 5 4 3 2 1

    List of Contributors

    Preface

    The aim of this handbook is to provide readers with an overview of cutting-edge research on the dynamics and evolution of financial markets. While the insights offered in this book will be valuable for the future development of finance theory, we are convinced they are also of vital importance to today's financial practitioners. All chapters are written exclusively for this handbook with the goal of being accessible to the nonspecialist reader, may they be asset managers or researchers from other disciplines.

    The view of financial markets promoted here goes far beyond traditional finance approaches to asset management. The classic credo is still to buy and hold a market portfolio or, in more sophisticated versions, to place bets on the convergence of asset prices to some equilibrium. In contrast, the models presented in this book aim to explain the market dynamics of asset prices based on the heterogeneity of investors. This can offer insights for asset management approaches including market timing, which is potentially very fruitful but also very difficult without a clear understanding of the various interactions in a financial market.

    Although this handbook is not the only work in finance highlighting the importance of dynamics and heterogeneity for financial markets, it is unique because it is the most recent and most encompassing account of this literature. Other important contributions to the general theme are, for example, Shefrin's excellent book, A Behavioral Approach to Asset Pricing, and Volume 2 of the Handbook of Computational Economics edited by Tesfatsion and Judd, both published by Elsevier. As compared to this one, the two other contributions have a different focus; however, Shefrin's book is less dynamic because it is fully based on general equilibrium, and the Tesfatsion and Judd chapters of the book dealing with finance focus more on illustrating the dynamics of heterogeneous agents models by computational simulations.

    The importance of this work for the development of finance theory is best explained by contrasting it to the main paradigm in finance: optimization and rational expectations as theoretical underpinnings of the efficient market hypothesis. The prevalent view of traditional finance is that of any point in time all traders make use of all available information; and as a consequence, any predictable pattern, such as a price trend must already be anticipated and reflected in current prices. Only the arrival of new information can lead to price changes. In 1964 Cootner formulated the conjecture that period-by-period price changes are random movements statistically independent of each other. This stochastic price mechanism is at the heart of many of the key theoretical models in finance such as optimal portfolio rules inspired by the work of Markowitz and Merton from 1952 onward; the static and intertemporal capital asset pricing models of Sharpe, Lintner, Mossin, and Merton from the 1960s; and models for the pricing of contingent claims beginning in the 1970s with the work of Black and Scholes.

    The two main theoretical justifications of the traditional finance view—optimization and rational expectations—have been under heavy attack for some time and will clearly not emerge unscathed. One may argue that people think twice when money is involved, coming to a conclusion that is void of any biases or mistakes. Empirical evidence for this view is weak at best. To the contrary, high monetary gains (or losses) are often observed to trigger emotions that severely distort traders' decisions. The second argument is that the market itself will take care of irrational behavior and erase it through the force of market selection. This conjecture—made by Cootner, Friedman, and Fama—is challenged on the basis of theoretical and practical work (see, for example, Blume and Easley's contribution, Market Competition and Selection, to the New Palgrave Dictionary of Economics). Work by Shleifer and others has shown that too many irrational investors are a risk to rational investors because they cannot be arbitraged away, at least in the short run. One of the contributions of this handbook is to show the state of the current debate on the market-selection hypothesis—a debate that still has not come to a definite conclusion.

    Recent empirical and experimental work challenged the traditional view of efficient markets and the long-sustained belief in market rationality; see, for example, the excellent surveys on asset pricing in the Journal of Finance by Campbell (2000) and Hirshleifer (2001). Indeed a new paradigm based on behavioral models of decision under risk and uncertainty is beginning to crowd out the traditional view based on complete rationality of all market participants. The traditional and the behavioral finance models, however, share one important feature: They are both based on the notion of a representative agent—although this mythological figure is dressed differently. While traditionally he had rational preferences, expectations, and beliefs, he is currently a prospect theory maximizer, unable to carry out Bayesian updating and likely to fall into framing traps.

    The chapters in this book, in contrast, suggest models of portfolio selection and asset price dynamics that are explicitly based on the idea of heterogeneity of investors. They are descriptive and normative as well, answering which set of strategies one would expect to be present in a market and how to find the best response to any such market. The models presented are successful as a descriptive approach because they are able to explain facts of asset prices such as fat tails in the return distribution, stochastic and clustered volatility, and bubbles and crashes—facts that are anomalies or puzzles in the traditional finance world. On the second issue, the main observation is that there is nothing like the best strategy because the performance of any strategy will depend on all strategies in the market. Rationality therefore is to be seen as conditional on the market ecology. The key to investment success, thus, is understanding the interaction of the various strategies.

    This handbook has nine chapters on topics within the emerging field of dynamics and evolution in financial markets. They aim to explore the preceding ideas in consistent and adequate models with the goal of contributing to a better understanding of the dynamics of financial markets. The collection of chapters reflects the diversity of evolutionary approaches in terms of both conceptual and methodological aspects. On the conceptual level, readers will be exposed to several different modeling approaches: temporary and general equilibrium models are considered; dynamic systems theory, as well as game-theoretic reasoning, is applied; traders' behavior originates from expected utility maximization, genetic learning, or is only restricted by being adapted to the information filtration; and fundamentalists and noise traders also enter the stage. On the methodological level, readers will see analytical, empirical, and numerical techniques applied by this book's authors. In the best tradition of the Handbooks in Finance series, all chapters share a thorough and formal treatment of the issue under consideration. As with every growing area of research, we expect to see further progress and fruitful applications in this exciting field.

    Chapter 1, Thought and Behavior Contagion in Capital Markets by David Hirsh-leifer and Siew Hong Teoh, surveys more than 200 theoretical and empirical papers that emphasize the social interaction of traders. The authors argue that the analysis of thought contagion and the evolution of financial ideologies, and their effects on markets, is a missing chapter in modern finance, including behavioral finance. While financial practitioners always emphasize that their decisions are influenced not only by fundamentals and price movements but also by opinions expressed (e.g., in the media), theorists have done little to provide them with models that can check the consistency of these claims, and moreover help them to better understand in which direction financial markets might move. Given the progress in information technology that allows researchers to categorize and to rapidly put into context any piece of news, we can expect profitable trading strategies to evolve from the novel research on behavior contagion in capital markets.

    Chapter 2, How Markets Slowly Digest Changes in Supply and Demand by Jean-Philippe Bouchaud, J. Doyne Farmer, and Fabrizio Lillo, is a beautiful piece on the market microstructure of financial markets that makes the econophysics approach accessible to a wide audience. It exemplifies how natural scientists do research: In contrast to economics and finance, observations are more important than theories. It is argued convincingly that the standard dichotomy of finance between informed and uninformed traders can neither be supported empirically nor is it useful theoretically because it leads to overly complicated models. The authors suggest, instead, distinguishing between speculators and liquidity traders. They develop a theory of market liquidity and show that block trades can be traced back in markets for several days. Moreover, this chapter's authors have proven in practice that their insights are very valuable—measured in real money.

    Chapter 3, Stochastic Behavioral Asset-Pricing Models and the Stylized Facts by Thomas Lux, acknowledges that traditional finance in the form of the efficient market hypothesis still plays a dominant role in explaining the first moment of asset returns by the martingale property, but that it fails to explain robust stylized facts concerning higher moments such as fat tails of the return distribution and stochastic and clustered volatility. The chapter outlines recent models of stochastic interaction of traders using simple behavioral rules that can explain these stylized facts as emergent properties of interactions and dispersed activities of a large ensemble of agents populating the market place. Understanding the properties of higher moments of asset returns is very profitable, since using derivatives allows the exploitation of predictability of any degree. Moreover, showing that market behavior emerges as a fundamentally different behavior than individual behavior avoids wasting money on simple analogies often used in standard finance, such as the representative agent, according to which market behavior is of the same type as individual behavior.

    Chapter 4, Complex Evolutionary Systems in Behavioral Finance by Cars Hommes and Florian Wagener, provides inspiring theoretical, empirical and experimental results. The theoretical results are outlined by a simple adaptive beliefs system based on trading strategies derived from mean-variance analysis with heterogeneous beliefs. In the most simple setting, beliefs are of two types: fundamentalists and trend followers. The population weights are driven by the success of the strategies. The model results range from perfect foresight equilibria to chaotic dynamics. This model has become the leading paradigm of heterogeneous agents models, and it has been generalized in many directions, one of which is the large type limit case—an approximation of a market with many different trader types—which is also outlined in this chapter. The model does well on yearly data going back to 1871. Finally, the main results of the model are validated by forecasting experiments.

    Chapter 5, Heterogeneity, Market Mechanism, and Asset Price Dynamics by Carl Chiarella, Roberto Dieci, and Xue-Zhong He, shows the similarities and the differences in market dynamics between models populated by constant absolute and constant relative risk-averse agents. In both cases, agents have heterogeneous price expectations, some of which are formed by simple rules of thumb, and markets clear through a Walrasian auctioneer or through a market maker. The resulting dynamics are nonlinear and stochastic and differ according to the market-clearing mechanism assumed. As the authors show empirically, the framework nicely explains asset-price features such as fat tail behavior, volatility clustering, power-law behavior in returns, and bubbles and crashes.

    Chapter 6, Perfect Forecasting, Behavioral Heterogeneities, and Asset Prices by Jan Wenzelburger, develops an intertemporal CAPM with heterogeneous expectations that lies between models in which agents have perfect foresight and models with exogenous and ad hoc expectation rules. The twist of this contribution is to properly define the expectations of rational agents, which consider that other agents may be irrational and that their own behavior has some market impact. Such perfect forecasts need to solve for the temporary equilibria of each period. These issues are discussed for the CAPM, with detailed derivation of the asset-price dynamics and conditions for market selection and survival of agents using various expectations.

    Chapter 7, Market Selection and Asset Pricing by Lawrence Blume and David Easley, focuses on the old but nevertheless unsettled and very important debate about whether markets select for rational agents. The framework is traditional in its choice of a dynamic general equilibrium model populated by infinitely lived subjective expected utility maximizers. This chapter provides a very important link between finance and mainstream economics because the main topics of this handbook are exemplified in a model setup that every classically trained finance or economics student knows by heart. The main result discussed is that if equilibrium allocations are Pareto-efficient, markets select for rational agents (i.e., the market selects for those traders whose subjective beliefs are closest to the objective probabilities with which the states of the world occur). The chapter also provides insights into the deeper workings of market selection in this modeling framework by describing the discipline imposed by the market. A relevant and interesting issue is the analysis of the relationship between market selection and the noise trader literature.

    Chapter 8, Rational Diverse Beliefs and Market Volatility by Mordecai Kurz, outlines a model in which all market participants do the very best they can—but not better. Every agent keeps track of all publicly available information and builds expectations that are consistent with this observation. This modeling approach leaves sufficient heterogeneity to explain asset-price features that, according to the rational expectations literature, are anomalous or puzzling: excess volatility of asset returns, high- and time-varying risk premia, high volume of trade, and so on. Rational diverse beliefs turn out to provide a realistic and flexible paradigm between the two extremes—rational expectations and arbitrary ad hoc beliefs.

    Chapter 9, Evolutionary Finance by Igor V. Evstigneev, Thorsten Hens, and Klaus Reiner Schenk-Hoppe, studies market selection among traders following behavioral rules that may not necessarily be generated by utility maximization. The model allows for complete and incomplete markets and for short-and long-lived assets. The surprising finding of the literature surveyed is that, even though the pool of behavioral rules is quite large and the model is fairly general, a simple fundamental trading strategy—investing proportional to the expected relative dividends—does the trick (i.e., achieves the highest expected growth rate) or is at least the unique evolutionary stable trading strategy. This trading strategy can be seen as the Kelly Rule—betting your beliefs—applied in a market that generates returns endogenously from the interaction of trading strategies. One possible application of this result is to explain the success of value investing. Moreover, aggressive betting strategies in markets with endogenous odds, such as stock markets, can be derived from these results.

    This handbook's nine chapters can be characterized by several attributes, one of which is the time scale of the dynamics studied. Chapters 6 and 8 study expectation dynamics (i.e., the adjustment and learning process of boundedly rational investors). For most investors these dynamics happen on a medium time scale (months, quarters, even years). Both chapters develop new expectation hypotheses that are somewhere in between simple ad hoc heuristics and rational expectations. This interpretation can also be given for Chapter 4, in particular since the asset-pricing application is based on annual data.

    Chapter 2 goes to a much smaller time scale: intraday dynamics where the market microstructure—and in particular the market-clearing mechanism—plays a crucial role. Chapter 5 is concerned with these issues as well. Chapter 3 is somewhere in between the high-frequency intraday scale and medium-term dynamics, as can be seen from the attempt to explain daily return data. Finally, Chapters 7 and 9 consider long-term dynamics because they study market selection determined by the evolution of wealth. Chapter 1 surveys models across the board.

    Alternatively, the nine chapters can also be ordered according to the degree of rationality of the traders considered. Chapter 7 is closest to the traditional view of complete rationality since agents maximize subjective expected utility and have correct price expectations. Chapters 6 and 8 define notions of rationality that are still quite demanding but more realistic: in Chapter 6 the problem of what a completely rational agent should expect in a market with irrational agents is solved, while Chapter 8 defines a notion of rationality that uses all available information but not more than that. Further down the road to a smaller degree of rationality, we find the modeling approach that Chapters 1 to 5 Chapter 2 Chapter 3 Chapter 4 Chapter 5 outline. Agents maximize but they may not have completely rational price expectations. Finally, the approach of Chapter 9 dismisses all assumptions on rationality by moving to a purely behavioral model of investment.

    It is our hope that this handbook, which encompasses several directions of current developments in dynamic and evolutionary models of financial markets, will serve interested readers by providing insight and inspiration.

    Introduction to the Series

    Advisory Editors

    Kenneth J. Arrow, Stanford University; George C. Constantinides, University of Chicago; B. Espen Eckbo, Dartmouth College; Harry M. Markowitz, University of California, San Diego; Robert C. Merton, Harvard University; Stewart C. Myers, Massachusetts Institute of Technology; Paul A. Samuelson, Massachusetts Institute of Technology; and William F. Sharpe, Stanford University.

    The Handbooks in Finance are intended to be a definitive source for comprehensive and accessible information. Each volume in the series presents an accurate, self-contained survey of a subfield of finance, suitable for use by finance and economics professors and lecturers, professional researchers, and graduate students and as a teaching supplement. The goal is to have a broad group of outstanding volumes in various areas of finance.

    Chapter 1. Thought and Behavior Contagion in Capital Markets

    Introduction 2

    Sources of Behavioral Convergence 5

    Rational Learning and Information Cascades: Basic Implications 7

    What Is Communicated or Observed? 9

    Observation of Past Actions Only10

    Observation of Consequences of Past Actions15

    Conversation, Media, and Advertising16

    Psychological Bias 17

    Reputation, Contracts, and Herding 18

    Security Analysis 20

    Investigative Herding20

    Herd Behavior by Stock Analysts and Other Forecasters21

    Herd Behavior and Cascades in Security Trading 24

    Evidence on Herding in Securities Trades24

    Financial Market Runs and Contagion27

    Exploiting Herding and Cascades28

    Markets, Equilibrium Prices, and Bubbles 29

    Cascades and Herding in Firm Behavior 36

    Investment and Financing Decisions36

    Disclosure and Reporting Decisions38

    Contagion of Financial Memes 39

    Conclusion 44

    References46

    [intro]

    Abstract

    Prevailing models of capital markets capture a limited form of social influence and information transmission, in which the beliefs and behavior of an investor affect others only through market price, information transmission and processing is simple (without thoughts and feelings), and there is no localization in the influence of an investor on others. In reality, individuals often process verbal arguments obtained in conversation or from media presentations and observe the behavior of others. We review here evidence about how these activities cause beliefs and behaviors to spread and affect financial decisions and market prices; we also review theoretical models of social influence and its effects on capital markets. To reflect how information and investor sentiment are transmitted, thought and behavior contagion should be incorporated into the theory of capital markets.

    Note: We thank Jason Chan, SuJung Choi, and Major Coleman for their valuable research assistance.

    Keywords: capital marketsthought contagionbehavioral contagionherd behaviorinformation cascadessocial learninginvestor psychologyaccounting regulationdisclosure policybehavioral financemarket efficiencypopular modelsmemes

    1.1.. INTRODUCTION

    The theory of capital market trading and pricing generally incorporates only a limited form of social interaction and information transmission, wherein the beliefs and behavior of an investor affect other investors only through market price. Furthermore, in standard capital market models, there is no localized contagion in beliefs and trading. Trading behaviors do not move from one investor to other investors who are proximate (geographically, socially, professionally, or attentionally through connectivity in the news media). Even most recent models of herding and information cascades in securities markets involve contagion mediated by market price so that there are no networks of social interaction. Furthermore, existing behavioral models of capital market equilibrium do not examine how investors form naïve popular ideas about how capital markets work and what investors should do, and how such popular viewpoints spread.[¹]

    ¹Section 1.9 discusses work on social networks and securities trading (DeMarzo, Vayanos, and Zwiebel, 2001, and Ozsoylev, 2005) and learning in standard capital market models. The work of Robert Shiller and coauthors on popular models in finance is discussed in Section 1.11.

    The theory of investment has incorporated social interactions somewhat more extensively, both in the analysis of increasing returns and path dependence (see Arthur, 1989) and in models of social learning about the quality of investment projects (discussed in Section 1.10.1). However, traditional models of corporate investment decisions do not examine the process of contagion among managers of ideas about investment, financing, disclosure, and corporate strategy.

    In reality, individuals often observe others' behavior and obtain information and ideas through conversation and through print and electronic media. Individuals process this information through both reasoning and emotional reactions rather than performing the simple Bayesian or quasi-Bayesian updating of standard rational or behavioral models. Popular opinions about investment strategies and corporate policies evolve over time, partly in response to improvements in scientific understanding and partly as a result of psychological biases and other social processes. We are influenced by others in almost every activity, and price is just one channel of influence. Such influence can occur through rational learning (see, e.g., Banerjee, 1992; Bikhchandani, Hirshleifer, and Welch, 1992) or through through irrational mechanisms (see Section 1.5), the latter including the urge to conform (or deviate) and contagious emotional responses to stressful events.

    This essay reviews theory and evidence about the ways beliefs about and behaviors in capital markets spread. We consider here decisions by investors about whether to participate in the stock market and what stocks to buy; decisions by managers about investment, financing, reporting, and disclosure; and decisions by analysts and media commentators about what stocks to follow, what stocks to recommend, and what forecasts to make. We also consider the effects of contagion on market prices, regulation, and welfare as well as policy implications.

    We argue that in actual capital markets, in addition to learning from price, a more personal form of learning is also important: from quantities (individual actions), from performance outcomes, and from conversation—which conveys private information, ideas about specific assets, and ideas about how capital markets work. Furthermore, we argue that learning is often local: People learn more from others who are proximate, either geographically or through professional or other social networks. We therefore argue that social influence is central to economics and finance and that contagion should be incorporated into the theory of capital markets.

    Several phenomena are often adduced as evidence of irrational conformism in capital markets, such as anecdotes of market price movements without obvious justifying news; valuations which, with the benefit of hindsight, seem like mistakes (such as the valuations of U.S. Internet stocks in the late 1990s or of mortgage-backed securities in recent years); the fact that financial activity such as new issues, IPOs, venture capital financing, and takeovers move in general or sector-specific waves (see, e.g., Ritter and Welch, 2002; Rau and Stouraitis, 2008). Observers are often very quick to denounce alleged market blunders and conclude that investors or managers have succumbed to contagious folly.

    There are two problems with such casual interpretations. First, sudden shifts do not prove that there was a blunder. Large price or quantity movements may be responses to news about important market forces. Second, even rational social processes can lead to dysfunctional social outcomes.

    With respect to the first point, market efficiency is entirely compatible with massive ex post errors in analyst forecasts and market prices and with waves in corporate transaction actions in response to common shifts in fundamental conditions.

    With respect to the second point, the theory of information cascades (defined in Section 1.2) and rational observational learning shows that some phenomena that seem irrational can actually arise naturally in fully rational settings. Such phenomena include (1) frequent convergence by individuals or firms on mistaken actions based on little investigation and little justifying information; (2) fragility of social outcomes with respect to seemingly small shocks; and (3) the tendency for individuals or firms to delay decision for extended periods of time and then, without substantial external trigger, suddenly to act simultaneously. Furthermore, theoretical work has shown that reputation-building incentives on the part of managers can cause convergent behavior (Item 1) and has also offered explanations for why some managers may deviate from the herd as well. So care is needed in attributing either corporate event clustering or large asset price fluctuations to contagion of irrational errors.[²]

    ² Recent reviews of theory and evidence of both rational observational learning and other sources of behavioral convergence in finance include Devenow and Welch (1996), Hirshleifer (2001), Bikhchandani and Sharma (2001), and Daniel, Hirshleifer, and Teoh (2002).

    In addition to addressing these issues, we consider a shift in analytical point of view from the individual to the financial idea or meme. A meme, first defined by Dawkins (1976), is a mental representation (such as an idea, proposition, or catchphrase) that can be passed from person to person. Memes are therefore units of cultural replication, analogous to the gene as a unit of biological heredity. The field of memetics views cultural units as replicators, which are selected upon and change in frequency within the population. Just as changes in gene frequency imply evolution within biologically reproducing populations, changes in meme frequency imply cultural evolution. We argue that certain investment theories have properties that make them better at replicating (more contagious or more persistent), leading to their spread and survival.

    Furthermore, we argue that through cumulative evolution, financial memes combine into coadapted assemblies that are more effective at replicating their constituent memes than when the components operate separately. We call these assemblies financial ideologies. Memetics offers an intriguing analytical approach to understanding the evolution of capital market (and other) popular beliefs and ideologies.

    Only a few finance scholars have emphasized the importance of popular ideas about markets (especially Robert Shiller, as mentioned in Footnote 1), and there has been very little formal analysis of the effects and spread of popular financial ideas. We argue here that the analysis of thought contagion and the evolution of financial ideologies, as well as their effects on markets, constitute a missing chapter in modern finance, including behavioral finance.

    Our focus is on contagion of beliefs or behavior rather than defining contagion as occurring whenever one party's payoff outcomes affect another's. Therefore we do not review systematically the literature on contagion in bankruptcies or international crises in which fundamental shocks and financial constraints cause news about one firm or region to affect the payoffs of another.

    Section 1.2 discusses learning and the general sources of behavioral convergence. Section 1.3 discusses basic implications of rational learning and information cascades. Section 1.4 discusses basic principles of rational learning models and alternative scenarios of information transfer by communication or observation. Section 1.5 examines psychological bias and herding. Section 1.6 describes agency and reputation-based herding models. Section 1.7 describes theory and evidence on herding and cascades in security analysis. Section 1.8 describes herd behavior and cascades in security trading. Section 1.9 describes the price implications of herding and cascading. Section 1.10 discusses herd behavior and cascading in firms' investment, financing, and disclosure decisions. Section 1.11 examines the popular models or memes about financial markets. Section 1.12 concludes.

    1.2.. SOURCES OF BEHAVIORAL CONVERGENCE

    An individual's thoughts, feelings, and actions are influenced by other individuals by several means: verbal communication, observation of actions (e.g., quantities such as supplies and demands), and observation of the consequences of actions (such as payoff outcomes or market prices). Our interest is in convergence or divergence brought about by direct or indirect social interactions (herding or dispersing). So we do not count random groupings that arise solely by chance as herding, nor do we count mere clustering, wherein individuals act in a similar way owing to the parallel independent influence of a common external factor.

    Following Hirshleifer and Teoh (2003a) we define herding/dispersing as any behavior similarity or dissimilarity brought about by the direct or indirect interaction of individuals.[³] Possible sources include the following:

    Payoff externalities (often called strategic complementarities or network externalities). For example, there is little point to participating in Facebook unless many other individuals do so as well.

    Sanctions upon deviants. For example, critics of a dictatorial regime are often punished.

    Preference interactions. For example, a teenager may want an iPhone mainly because others talk about the product, though a few mavericks may dislike a product for the same reason.

    Direct communication. This is simply telling; however, just telling often lacks credibility.

    Observational influence. This is an informational effect wherein an individual observes and draws inferences from the actions of others or the consequences of those actions.

    ³ The interaction required in our definition of herding can be indirect. It includes a situation in which the action of an individual affects the world in a way that makes it more advantageous for another individual to take the same action, even if the two individuals have never directly communicated. But mere clustering is ruled out.

    We can distinguish an informational hierarchy and a payoff hierarchy in sources of convergence or divergence (see also Hirshleifer and Teoh, 2003a). The most inclusive category, herding/dispersing, includes both informational and payoff interaction sources of herding as special cases.

    Within herding/dispersing, the informational hierarchy is topped by observational influence, a dependence of behavior on the observed behavior of others or the results of their behavior. This influence may be either rational or irrational. A subcategory is rational observational learning, which results from rational Bayesian inference from information reflected in the behavior of others or the results of their behavior. A further refinement of this subcategory consists of information cascades, wherein the observation of others (their actions, payoffs, or statements) is so informative that an individual's action does not depend on his own private signal.[⁴]

    ⁴ See Bikhchandani, Hirshleifer, and Welch (1992); Welch (1992). Banerjee (1992) uses a different terminology for this phenomenon.

    Imitation, broadly construed, includes both information cascades and subrational mechanisms that produce conformity with the behavior of others. A crucial benefit of imitation is the exploitation of information possessed by others. When an insider is buying, it may be profitable to buy even without knowing the detailed reason for the purchase. There is also contagion in the emotions of interacting individuals (see, e.g., Barsade, 2002). The benefits of imitation are so fundamental that the propensity to follow the behaviors of others has evolved by natural selection. Imitation has been extensively documented in many animal species, both in the wild and experimentally.[⁵]

    ⁵ See, e.g., Gibson and Hoglund (1992), Giraldeau (1997), and Dugatkin (1992). Some authors use definitions of imitation that require substantial understanding on the part of the imitator, in which case such imitation is rare among nonhumans.

    In an information cascade, since an individual's action choice does not depend on his signal, his action is uninformative to later observers. Thus, cascades are associated with information blockages (Banerjee, 1992; Bikhchandani, Hirshleifer, and Welch, 1992), and, as we will see, with fragility of decisions (Bikhchandani, Hirshleifer, and Welch, 1992). Information blockages are caused by an informational externality: An individual chooses her actions for private purposes, with little regard for the potential information benefit to others.[⁶]

    ⁶ Chamley (2004b) and Gale (1996) review models of social learning and herding in general. For presentation of information cascades theory and discussion of applications, tests, and extensions, see Bikhchandani, Hirshleifer, and Welch (1998, 2008a). Bikhchandani, Hirshleifer, and Welch (2008b) provides an annotated bibliography of research relating to cascades.

    A payoff interaction hierarchy provides a distinct hierarchy of types of herding or dispersing that intersects with the categories in the information hierarchy. The first subcategory of the catch-all herding/dispersing category is payoff and network externalities. This consists of behavioral convergence or divergence arising from the effects of an individual's actions on the payoffs to others of taking that action. Direct payoff externalities have been proposed as an explanation for bank runs (Chari and Jagannathan, 1988; Diamond and Dybvig, 1983), since a depositor who expects other depositors to withdraw has a stronger incentive to withdraw, and clumping of stock trades by time (Admati and Pfleiderer, 1988) or exchange (Chowdhry and Nanda, 1991), since uninformed investors have an incentive to try to trade with each other instead of with the informed.

    In several models, a desire for good reputation causes payoffs to depend on whether individual behaviors converge.[⁷] Thus, a subcategory of the payoff and network externalities category is reputational herding and dispersion, wherein behavior converges or diverges owing to the incentive for a manager to maintain a good reputation with some observer. When individuals care about their reputations, reputational herding and information cascades can both easily occur, since an individual who seeks to build a reputation as a good decision maker may rely on the information of earlier decision makers (Ottaviani and Sørensen, 2000).

    ⁷ See Scharfstein and Stein (1990), Rajan (1994), Trueman (1994), Brandenburger and Polak (1996), Zwiebel (1995), and Ottaviani and Sørensen (2006).

    1.3.. RATIONAL LEARNING AND INFORMATION CASCADES: BASIC IMPLICATIONS

    If many individuals possess conditionally independent signals about which choice alternative is better, their information could be aggregated to determine the right decision with arbitrarily high precision. Information cascades lead to information blockage, which reduces the quality of later decisions. This blockage also has several other implications for the contagion and stability of financial decisions, some of which hold even in rational learning settings in which cascades proper do not occur.

    Consider a sequence of individuals who face ex ante identical choices (e.g., investment projects), observe conditionally independent and identically distributed private information signals, and observe the actions but not the payoffs of predecessors. Suppose that individual i is in a cascade and that later individuals understand this. Then individual i + 1, having learned nothing from the choice of i, is in an informationally identical position to that of i. So i + 1 also makes the same choice regardless of his private signal. By induction, this reasoning extends to all later individuals; the pool of information implicit in the past actions of individuals stops growing when a cascade begins. Indeed, in the simplest possible cascades setting, at this point the quality of decisions never improves again.

    When the assumptions are modified slightly, information is not blocked forever. If individuals are not identical ex ante, then the arrival of an individual with deviant information or preferences can dislodge a cascade. For example, an individual with a highly precise signal will act independently, which conveys new information to later individuals. Furthermore, the arrival of public news, either spontaneously and independently of past choices or as payoff outcomes from past choices, can dislodge a cascade. The more generic implication of the cascades approach is that the quality of decisions improves much more slowly than would be the case under ideal information aggregation. Information blockages can last for substantial periods of time; as we will see, at such times social outcomes are often fragile.

    Information cascades are a special case of behavioral coarsening, defined as any situation in which an individual takes the same action for multiple signal values. When there is behavioral coarsening, as in an information cascade, an individual's action does not fully convey his information signal to observers. So, where a cascade causes (at least temporarily) a complete information blockage, behavioral coarsening leads to partial blockage. A surprising aspect of the theory of information cascades is that in a natural setting the most extreme form of behavioral coarsening occurs.

    Since information is aggregated poorly in an information cascade, the quality of decisions is reduced. Rational individuals who are in a cascade understand that the public pool of information implicit in predecessors' actions is not very precise. As a result, even a rather small nudge, such as a minor public information disclosure, can cause a well-established and thoroughly conventional behavior pattern to switch.

    The arrival of a meaningful but inconclusive public information disclosure can, paradoxically, reduce the average quality of individuals' decisions. Other things can equal, a given individual is better off receiving the extra information in the disclosure. However, additional information will sometimes cause individuals to cascade earlier, aggregating the information of fewer individuals. On balance, the public signal can induce a less informative cascade (Bikhchandani, Hirshleifer, and Welch, 1992). Of course, if highly conclusive public information arrives, rational individuals will make very accurate decisions.

    The dangers of a little learning are created in other information environments as well. In cascade models, the ability of individuals to observe payoff outcomes in addition to past actions, or to more precisely make a noisy observation of past actions can reduce the average accuracy of decisions (Cao and Hirshleifer, 1997, 2002). Also, the ability to learn by observing predecessors can make the decisions of followers noisier by reducing their incentives to collect (perhaps more accurate) information themselves (Cao and Hirshleifer, 1997). Furthermore, even if an unlimited number of payoff outcome signals arrive, the choices that individuals can make may limit the resulting improvement in the information pool. For example, there can be a positive probability that a mistaken cascade will last forever (Cao and Hirshleifer, 2002).

    Often individuals choose not only whether to adopt or reject a project but when to do so. As a result, the timing and order of moves, which are given in the basic cascade model, are endogenously determined. In models of the option to delay investment choices,[⁸] there can be long periods with no investment, followed by sudden spasms in which the adoption of the project by one firm triggers investment by others.

    ⁸ See Chamley and Gale (1994); see also Hendricks and Kovenock (1989); Bhattacharya, Chatterjee, and Samuelson (1986); Zhang (1997) and Grenadier (1999); and Chamley (2004a, 2004b).

    Most of the conclusions described here generalize to other social learning settings in which cascades proper do not occur. Even when information blockage is not complete, information aggregation is limited by the fact that individuals privately optimize rather than taking into account their effects on the public information pool. This creates a general tendency for information aggregation to be self-limiting. At first, when the public pool of information is very uninformative, actions are highly sensitive to private signals, so actions add a lot of information to the public pool.[⁹] As the public pool of information grows, individuals' actions become less sensitive to private signals.

    ⁹ The addition can be directly through observation of past actions or indirectly through observation of consequences of past actions, as in public payoff information that results from new experimentation with various choice alternatives.

    In the simplest versions of the cascade model, behavioral coarsening occurs in an all-or-nothing fashion so that there is either full use of private signals or no use of private signals (as in Banerjee, 1992, and the binary example of Bikhchandani, Hirshleifer, and Welch, 1992). In more general settings, coarsening occurs by degrees, but complete blockage eventually occurs (see the cascades model with multiple signal values of Bikhchandani, Hirshleifer, and Welch, 1992). In some settings, coarsening can gradually proceed without ever reaching a point of complete blockage, though the probability that an individual uses his own signal asymptotes toward zero, a phenomenon called limit cascades (Smith and Sørensen, 2000). Or, if there is observation noise, the public pool of information can grow steadily but more and more slowly (Vives, 1993).

    So whether information channels become gradually or quickly clogged and whether the blockage is partial or complete depends on the economic setting, but the general conclusion that there can be long periods in which individuals herd upon poor decisions is robust. In addition, there tends to be too much copying or behavioral convergence; someone who uses her own private information heavily provides a positive externality to followers, who can draw inferences from her action.

    Information cascades result from the individual's private signal being overwhelmed by the growing public pool of information. Such an outcome is impossible in a setting where there is always a chance that an individual will receive a signal that is conclusive or arbitrarily close to conclusive. However, if near-conclusive signals are rare, the public information pool can grow very slowly, in which case information cascade can be a good approximation. Indeed, as the quality of the public information pool improves, the likelihood that an individual will receive a signal powerful enough to oppose it declines.

    To summarize, the information cascade model and some related rational learning theories provide a few key general implications. The first and central implication is idiosyncrasy, or poor information aggregation. Cascades tend to emerge rapidly, so the signals of a relatively small number of early individuals dominate the behavior of numerous followers.

    The second is fragility, or fads. The blockage of information aggregation that is characteristic of cascades makes behavior sensitive to small shocks. We are accustomed to thinking of sensitivity to shocks as a rare circumstance, as when a flipped coin lands on its side. The tendency of cascades to form suggests that real life is somewhat like Hollywood thrillers in which the chase scene inevitably ends with the hero's car teetering precariously at the edge of a precipice.

    The third is simultaneity, or delay followed by sudden joint action. Such effects are sometimes referred to as chain reactions, stampedes, or avalanches. Endogenous order of moves, heterogeneous preferences, and precisions can exacerbate these problems.

    The fourth is paradoxicality, or adverse effects on decision accuracy or welfare of informational improvements; the fifth is path dependence, or outcomes depending on the order of moves or signal arrival. This implication is shared with models of payoff interactions (e.g., Arthur, 1989).

    1.4.. WHAT IS COMMUNICATED OR OBSERVED?

    We now describe in somewhat more detail alternative sets of assumptions in observational influence models and the implications of these differences.

    1.4.1.. Observation of Past Actions Only

    Here we retain the assumption of the basic cascade model that only past actions are observable, but we consider several model variations.

    Discrete, Bounded, or Gapped Actions vs. Continuous Unbounded Actions

    If the action space is continuous, unbounded, and without gaps, an individual's action is always at least slightly sensitive to his private signal. Thus, actions always remain informative, and information cascades never form. So, for inefficient information cascades to occur, actions must be discrete, bounded, or gapped. As discrete or bounded action spaces become more extensive, cascades become more informative, approaching full revelation.[¹⁰]

    ¹⁰ See Lee (1993); see also Gul and Lundholm (1995) and Vives (1993) for continuous settings without cascades. Early cascade models were based on action discreteness (Bikhchandani, Hirshleifer, and Welch, 1992; Welch, 1992).

    The assumption that actions are discrete is often highly plausible. We vote for one candidate or another, not for a weighted average of the two. A worker is hired or not hired, fired or not fired. A takeover bidder either does or does not seek control of a target firm. Often alternative investment projects are mutually exclusive. Although the amount invested is often continuous, if there is a fixed cost the option of not investing at all is discretely different from positive investment.

    More broadly, one way in which the action set can be bounded is if there is a minimum and maximum feasible project scale. If so, when the public information pool is sufficiently favorable, a cascade at the maximum scale will form, and when the public information pool is sufficiently adverse, individuals will cascade on the minimum scale. Since there is always an option to reject a new project, investment has a natural extreme action of zero. Thus, a lower bound of zero on a continuous investment choice creates cascades of noninvestment (Chari and Kehoe, 2004). Similarly, gaps can create cascades.[¹¹]

    ¹¹ Asymmetry between adoption and rejection of projects is often realistic and has been incorporated in several social learning models of investment to generate interesting effects. As for gaps, sometimes either a substantial new investment, no change, or disinvestment is feasible, but fixed costs make a small change clearly unprofitable. If so, then a cascade upon no change is feasible. Similarly, a cascade of securities nontrading can form when there is a fixed cost of taking a long or short position, or when there is a minimum trade size. Even if the true action space is continuous, ungapped and unbounded, to the extent that observers are unable to perceive or recall small fractional differences, the actions of their predecessors effectively become either noisy or discrete.

    If perceptual discretizing is very finely grained, the outcome will still be very close to full revelation. However, perception and analysis are coarse; consider, for example, the tendency of people to round off numbers in memory and conversation. There is evidence of clustering for retail deposit interest rates around integers and that this is caused by limited recall of investors (Kahn, Pennacchi, and Sopranzetti, 1999).

    Observability of Predecessors' Payoffs or Signals

    Even if individuals observe a subset of past signals, such as the past k signals, since in general uncertainty remains, inefficient cascades can form. With regard to settings with observation of past payoffs, inefficient cascades can form and with positively probability last forever because a cascade can lock into an inferior choice before sufficient trials have been performed on the other alternative to persuade later individuals that this alternative is superior (Cao and Hirshleifer, 2002). We discuss research on the effects of observability of past payoffs and signals in more depth in Subsection 1.4.2.

    Costless vs. Costly Private Information Acquisition

    Individuals often expend resources to obtain signals, but they also often observe private signals costlessly in the ordinary course of life. Most social learning models take the costless route. Costs of obtaining signals can lead to little accumulation of information in the social pool for reasons similar to cascades or herding models with costless information acquisition. Individuals have less incentive to investigate or observe private signals if the primary benefit of using such signals is the information that such use will confer on later individuals. (Burguet and Vives, 2000, examine the conditions under which complete learning occurs in a continuum model with investigation costs.) Indeed, if the basic information cascade setting is modified to require individuals to pay a cost to obtain their private signals, once a cascade is about to start, an individual has no reason to investigate. The outcome is identical to the basic cascade model: information blockage. But the individual is acting without regard to her signal in only a degenerate sense: she has not acquired any signal to regard.

    This suggests an extended definition of cascades that can apply to situations in which private signals are costly to obtain. Following Hirshleifer and Teoh (2003a), we define an investigative cascade as a situation in which either:

    An individual acts without regard to his private signal, or

    An individual chooses not to acquire a costly signal, but he would have acted without regard to that signal had he been forced to acquire it at the same level of precision that he would have voluntarily acquired if he were unable to observe the actions or payoffs of others

    Item 1 implies that all information cascades are also investigative cascades. Item 2 is simplest in the special case of a binary decision of whether or not to acquire an information signal of exogenously given precision. Item 2 then reduces to the statement: The individual chooses not to acquire the signal, but if he were forced to acquire it he would ignore its realization (because of the information he has already gleaned by observing others).[¹²]

    ¹² Item 2 further allows for the purchase of different possible levels of precision. The definition focuses on the precision that the individual would select under informational autarky. If, under this precision, the individual's action does not depend on the realization, he is in an investigative cascade.

    Investigative cascades may occur in the decisions by individuals to invest in different countries. If investigation of each requires a fixed cost, then with a large number of countries investors may cascade on noninvestment (see the related analysis of Calvo and Mendoza, 2001).

    Observation of All Past Actions vs. a Subset or Statistical Summary of Actions

    Sometimes people can observe only the recent actions of others, a random sample of actions, or the behavior of neighbors in some geographic or other network.[¹³] In such settings mistaken cascades can still form. For example, if only the preceding k actions are observed, a cascade may form within the first k individuals and then through chaining extend indefinitely. Alternatively, individuals may only be able to observe a statistical summary of past actions. Information blockage and cascades are possible in such a setting as well (Bikhchandani, Hirshleifer, and Welch, 1992). A possible application is to the purchase of consumer products. Aggregate sales figures for a product matter to future buyers because they reveal how previous buyers viewed desirability of alternative products (Bikhchandani, Hirshleifer, and Welch, 1992; Caminal and Vives, 1999).

    ¹³Bala and Goyal (1998) analyze learning from the actions and payoff experiences of neighbors. They show that this leads to convergence of behavior and, under some conditions, efficient outcomes.

    Observation of Past Actions, Accurately or with Noise

    When past actions are observed with noise, social learning is still imperfect (Vives, 1993), and (depending on the setting) cascades can still form (Cao and Hirshleifer, 1997). In some scenarios a model in which individuals learn from price reduces, in effect, to a basic social learning model with indirect observation of a noisy statistical summary of the past trades of others.

    Choice of Timing of Moves vs. Exogenous Moves

    Consider a setting in which individuals or firms with private signals about project quality have a choice about whether to invest or delay. In other

    Enjoying the preview?
    Page 1 of 1